Abstract:
Currently scientists around the world are working to make new discoveries and advancements using e-Science platforms. Modern science relies on computational power and uses large data which often requires grid or utility computing. There are algorithms that analyse this large-scale data using significant quantities of computational power. Terabytes of data need to be mined for these algorithms to work effectively. e-Science provides scientists environment to work on this data using GUIs, workflow engines and other tools. The volumes of data that are becoming available for use in e-Science is growing exponentially and it will be very difficult in near future to process all of it in order to carry out research. It will therefore become extremely easy to miss important data which might yield useful findings. This thesis presents an approach which uses a MAS (Multi-agent Systems) to assist the researcher in carrying out their work and to manage large-scale data effectively. With current approaches the user runs a scientific workflow and then has to wait possibly several hours for output. They will need to repeat the entire process again if the results are not accurate, either by using new data or by changing algorithm parameters. Our proposed approach provides user a fairly generic way to automate this process. The user provides the system with the workflow and the kind of results that they are interested in using an intuitive interface. The MAS then goes through the data, selects the samples and passes them to the workflow execution environment. The user is notified of the desired results whenever they become available. In order to evaluate this approach a prototype has been implemented on a real-world bioinformatics case study. Experimental results have found that a multi-agent system can assist the user by saving both their time and effort in the analysis of large data sets. A researcher can assign agents to do basic repetitive tasks on their behalf which frees up their time and energy for more productive tasks.
